Stanford DeLM cuts multi-agent task costs 50%

Stanford researchers have developed DeLM, a decentralized language model framework that eliminates the need for a central orchestrator in multi-agent AI systems. This novel approach allows agents to coordinate directly, sharing verified progress and tasks asynchronously through a shared context and task queue. DeLM significantly reduces costs by approximately 50% and improves accuracy, outperforming traditional centralized systems on benchmarks like SWE-bench and LongBench-v2. The framework enables agents to build upon each other's findings, avoid redundant work, and focus on unresolved issues, making AI tasks more efficient and robust.